684 Numerical Evaluation of Dynamic Response Chap.

16

Explicit methods, typically the central difference method, are used in software such
as LS-DYNA, ABAQUS-Explicit, and OpenSees. Because the time step required is very
small, these methods are impractical for analysis of large systems using conventional com-
puters (one to four computer processors). However, explicit methods have the advantage
that they can be conveniently programmed for parallel computing using a large number of
computer processors. Most applications to large systems diagonalize the damping matrix
c so that k̂ is diagonal and the equations in step 2.2 can be solved efficiently. Researchers
have developed various approximate models for damping to achieve this goal.

16.3.2 Nonlinear Static Analysis

Nonlinear static analysis is used to investigate the force–deformation behavior of a struc-

ture for a specified distribution of forces, typically lateral forces. With certain assumption
of force distribution, nonlinear static analysis is called pushover analysis (see Chapters 19
and 22).
Dropping the inertia and damping terms in the equations of motion [Eq. (16.1.1)] gives
the system of nonlinear equations to be solved in a static problem:
f S (u) = p(t) (16.3.2)
Before examining multiple-force steps necessary in a pushover analysis, consider the so-
lution of the equilibrium equations for a single set of forces:
f S (u) = p (16.3.3)
The task is to determine the displacements u due to a set of given external forces p, where
the nonlinear force–deformation relation f S (u) is known for the system to be analyzed.
Suppose that after j cycles of iteration, u( j) is an estimate of the unknown displace-
ments and we are interested in developing an iterative procedure that provides an improved
( j+1)
estimate u( j+1) . Expanding the resisting forces f S (u) in Taylor series about the known
( j)
estimate u , and dropping the second- and higher-order terms, leads to the linearized
equation (see Section 5.7.1)
( j) ( j)
kT !u( j) = p − f S = R( j) (16.3.4)
where
!
( j) ∂f S (u) !!
kT = (16.3.5)
∂u !u( j)
is the tangent stiffness matrix at u( j) ; (kT ) i, j = the change in force at DOF i due to unit
change in displacement at DOF j at the current state of the system. Solving the linearized
system of equations (16.3.4) gives !u( j) and an improved estimate of displacements:
u( j+1) = u( j) + !u( j) (16.3.6)
This is the essence of the Newton–Raphson method for iterative solution of the nonlin-
ear equations (16.3.3). As shown in Section 5.7.1, this iterative method converges with
quadratic rate to the exact solution.
Sec. 16.3 Nonlinear Systems 685

The preceding description of the Newton–Raphson procedure for a single force step
can be generalized for multiple force steps. For this purpose, the forces are represented by
a reference spatial distribution pref and a variable scalar λi ; thus
pi = λi pref (16.3.7)
The nonlinear equilibrium equations for each force level are solved by Newton–Raphson
iteration starting with the initial estimate of the solution as the displacements at the previ-
ous force level. Table 16.3.2 summarizes such a procedure for nonlinear static analysis as
it might be implemented on the computer.
TABLE 16.3.2 NONLINEAR STATIC ANALYSIS

†
1.0 State determination for u = u0 : (f S )0 and (kT )0 .
2.0 Calculations for each force step, i = 0, 1, 2, . . .
( j) ( j) ( j)
2.1 Initialize j = 1, ui+1 = ui , (f S )i+1 = (f S )i , and (kT )i+1 = (kT )i .
2.2 pi+1 = λi+1 pref .
3.0 For each iteration, j = 1, 2, 3, . . .
( j) ( j)
3.1 Ri+1 = pi+1 − (f S )i+1 .
3.2 Check convergence; if the acceptance criteria are not met, implement steps 3.3
to 3.6; otherwise, skip these steps and go to step 4.0.
( j) ( j)
3.3 Solve (kT )i+1 !u( j) = Ri+1 ⇒ !u( j) .
( j+1) ( j)
3.4 ui+1 = ui+1 + !u( j) .
( j+1) ( j+1)
3.5 State determination: (f S )i+1 and (kT )i+1 .
3.6 Replace j by j + 1 and repeat steps 3.1 to 3.5; denote final value as ui+1 .
4.0 Repetition for next force step. Replace i by i + 1 and implement steps 2.0 and 3.0
for the next force step.
† u0 may be nonzero if initial gravity load effects are included in the analysis.

In step 3.2, the solution is checked and the iterative process is terminated when some
measure of the error in the solution falls below a specified tolerance. Typically, one or more
of the following convergence (or acceptance) criteria are enforced:
1. Residual force is less than a tolerance:
" ( j) "
"R " ≤ ε R (16.3.8a)
where $ · $ denotes the Euclidean norm of the vector. Conventional values for the
tolerance ε R range from 10−3 to 10−8 .
2. Change in displacement is less than a tolerance:
" ( j) "
"!u " ≤ εu (16.3.8b)
Conventional values for the tolerance εu range from 10−3 to 10−8 .
3. Incremental work done by the residual force acting through the change in displace-
ment is less than a tolerance:
" "
1 "
[!u( j) ]T R( j) " ≤ εw
2
(16.3.8c)
686 Numerical Evaluation of Dynamic Response Chap. 16

Tolerance εw must be at or near the computer (machine) tolerance because the left
side is a product of small quantities.
Although the examples presented subsequently use the preceding criteria, for
large MDF systems it is better to use relative force or displacement measures:
" ( j) " " ( j) "
"R " "!u "
% " ≤ εu%
≤ εR " (16.3.9a)
$pref $ "u( j) "

where the recommended value for tolerances ε%R and εu% is 10−3 to 10−6 . For frames,
the displacement vector contains translations and rotations (and the force vectors
contains forces and moments) whose magnitudes may be vastly different. For such
situations, we recommend use of relative incremental work to check convergence.
The convergence criterion then is
"# $T "
" "
" !u( j) R( j) "
"# $T " ≤ εw% (16.3.9b)
" "
" !u (1) R "
(1)

where the recommended value for εw% is on the order of 10−16 .

Example 16.2
The five-story shear building of Example 16.1 is subjected to monotonically increasing lateral
forces with the invariant distribution presented in Fig. E16.2a. The story shear–drift (Vj − δ j )
relationship is identical for all stories; it is bilinear with initial stiffness k = 100 kips/in.,
post yield stiffness ratio α = 0.05, and yield shear Vj y = 125 kips (Fig. E16.2b). Conduct
nonlinear static analysis of the building for the pref shown and force factors:
λT = & 0 1.0 1.1 1.2 1.3 1.4 1.5 1.6 '
Note that the base shear associated with pref is Vb = 125 kips, the yield value of base shear.
Vj , kips

625/15
k
500/15
k Vjy αk
375/15
k
250/15
k k
125/15
k δjy δj

(a) (b)

Figure E16.2 a, b